Disciplines such as machine learning, data science, and AI are heavily steeped in academia. However, they are rapidly showing their utility in creating real-world solutions to current economic, environmental, and social challenges by joining forces with economists, epidemiologists and other professionals. AI combined with IoT can help monitor wildfires in the Amazon. Modeling and neural networks can predict COVID-19 numbers. ML models can help support doctors in making discharge decisions.
Salesforce Research recently published The AI Economist, a research framework that studies how to improve economic design using AI to optimize productivity and social equality. This new open source AI framework for the economic policy was designed to simulate millions of years of economies – in parallel – to help economists, governments, and others design tax policies that optimize social outcomes in the real world.
Specifically, the AI Economist is a machine learning framework for economic design from Salesforce Research. It uses two-level multi-agent reinforcement learning (MARL) to train a “social planner” that sets taxes and subsidies in a simulated market economy that contains independently learning AI agents.
Why AI for tax policies?
Economic inequality is accelerating globally and is a growing concern due to its negative impact on economic opportunity, health, and social welfare. Many studies have shown that high-income inequality can negatively impact economic growth and economic opportunity. Taxes can help reduce inequality, but it is hard to find the optimal tax policy. Finding a tax policy that optimizes equality, along with productivity, is an unsolved problem. The AI Economist brings reinforcement learning (RL) to tax policy design for the first time to provide a purely simulation and data-driven solution.
Research has resulted in a model that is 16% more effective than a leading tax framework proposed by Emmanuel Saez. Instead, analysis has found that the AI Economist implements qualitatively different tax schedules than the baselines, with higher top tax rates and lower rates for middle incomes. It’s also been useful in simulations with human participants, achieving competitive equality-productivity trade-offs with the baselines and significantly higher income-weighted average social welfare. This suggests promise in using this approach to improve social outcomes in real economies.
Economic AI Simulation as a Learning Environment
The AI Economist is a purely simulation and data-driven approach to the design of optimal tax policies. It uses a principled economic simulation with both workers and a policy maker (a “planner” in the economics literature), collectively learning using reinforcement learning.
The simulation uses a two-dimensional world. There are two types of resources: wood and stone. Resources are scarce: they appear in the world at a limited rate. Workers move around, gather and trade resources, and earn income by building houses (this costs stone and wood). Houses block access: workers cannot move through the houses built by others. The simulation runs this economy throughout an episode, which is analogous to a “working career.”
Economic AI agents learned to “game” taxes to lower their effective tax rate. Every time the AI government intervenes or changes its tax policy, these agents are smart enough to find loopholes or ways to game the system, like alternating between tax periods with high and low incomes. From a technical point of view, that means things can become very unstable. Equality can fluctuate. Because AI agents are trying to maximize profits in a fluctuating environment, it means the simulation is closer to what it’s like in real life.
The AI Economist also achieves improvements in equality and productivity with real people earning real money, achieving significantly higher inverse-income weighted social welfare.
According to team member Nikhil Naik:
“Our model is incredibly powerful. Economists have previously relied on theorems, but theorems require simple math and are predicated on people behaving rationally. Our world today is getting more complex, and economic theories of the future need to incorporate additional requirements such as environmental protection seamlessly. Also, economic agents often exhibit complex, irrational, competitive, or collaborative behaviors. AI helps to model such complexity and a broad spectrum of behaviors.
Our model gives economists and policy-makers additional tools on which they can base their decisions. By simulating millions of years of economies and finding various tax frameworks, the AI Economist can predict how people would actually respond to a tax, like whether it will incentivize them to work more or work less.”
In September this year, The Salesforce Research Team launched an open source project to build an AI Economist for the real world. It aims to develop an AI social planner that optimizes economic policies and improves social welfare. They are seeking AI researchers, economists, and policy experts to help. You can view their documentation and signup to find out more.
Critical collaboration in response to social and economic challenges
Many startups and enterprises are applying AI to monitor, predict, and respond to complex global problems. An example is Omdena, a collaborative platform founded in May 2019, where a global community of changemakers builds AI solutions to real-world problems. There are currently 24 projects created across 84 countries, dealing with detecting wildfires, predicting climate change, and preventing gang violence.
Banking the unbanked
According to the World Bank, about 1.7 billion adults do not have an account at a financial institution or through a mobile money provider. One of the main reasons is that “first-time-borrowers” do not possess a credit history, collateral, or any previous accounts. All of which are essential for conventional credit scoring approaches.
45 researchers worked together in partnership with local financial institution Creedix, to build an ethical AI-based credit scoring system. Utilized, this would enable people to get access to fair and transparent credit. They determined the creditworthiness of an un-banked customer with alternate and traditional credit scoring data and methods, focusing initially on Indonesia, with three distinct data sets – with appropriate privacy safeguards:
The team decided on the following features taken from the project partner’s data and scrapped for additional data sources to fill in gaps. The scraped data came from numbeo to get the cost of living per area and living expenses. Data from Indeed provided salary numbers to assign an average salary to different kinds of jobs.
The final deliverables of the research project were engineered features and machine learning pipelines both for supervised and unsupervised learning. In the next step, the Creedix team will look through the features and decides on the weightage for each feature to build the final machine learning models.
Want to launch a developer event despite the challenges of COVID-19?
If you want to know more about how modern technologies and tools can support you for – and during – the organisation of a virtual event, don’t miss this article showcasing the best tools we used to host our online conferences since the COVID-19 outbreak.